Transfer results
How each chunk of data arrived: how fast it started, whether the gaps were even, whether it stopped midway, and whether it finished completely.
After you start, this lists the request, first data, every stall and the end, in order.
How the test works
Reproduces how AI services stream responses: one HTTPS connection stays open while the server pushes data on a fixed schedule.
The browser sends an HTTPS request to this site; the server replies in text/event-stream (SSE) format and keeps the connection open.
The server streams data for the selected duration (one chunk every 250 ms for Quick and Standard, every 500 ms for Deep). Each chunk carries a sequence number and server time, followed by an end marker.
The browser records when each chunk arrives and calculates time to first data, gaps, the longest stall and any missing sequence numbers to judge whether the transfer was complete.
This test
- Test tier
- Quick · 40 sec
- Simulated thinking
- Off
- Send interval
- 250 ms
- Chunks expected
- 160 chunks
- Chunks received
- Shown after the test
- Missing chunks
- Shown after the test
- Page moved to the background
- Shown after the test
- Tested at
- Shown after the test
Connection info
- Connection protocol
- Shown after the test
- Content type
- text/event-stream (SSE)
- Test endpoint
- This site's server /api/v1/stream
- Data format
- Each chunk carries a sequence number and server time
How we judge
A gap of more than 1.5 seconds between two chunks (6× the send interval) counts as a stall; disconnecting before the end marker counts as an early disconnect.
TTFBThe time from sending the request to receiving the first chunk. It includes the overhead of network round trips, proxy forwarding and connection setup.intervalThe time between two consecutive chunks. The server sends one every 250 ms, so on a smooth network the gaps are close to 250 ms too.max gapThe single longest wait during the test. Occasional brief hiccups are normal; only long stops hurt the reading experience.endA marker the server sends after the last chunk. Only when it arrives can we confirm the data was transferred completely and not cut off.How to reduce streaming interruptions
If the test shows stalls or early disconnects, check these three things. Exact names vary by proxy app and network.
The most common cause of stalls and disconnects is an unstable proxy node or route with heavy packet loss. Switch to another node, test again and compare.
Corporate gateways, HTTPS scanning in security software, or some proxy settings collect a streamed response into one block before handing it to the browser. It shows up as a long stall followed by everything arriving at once.
Browsers deprioritize background pages, and Wi-Fi switching or device sleep will also interrupt the connection. Keep the page open and your network stable during the test.
FAQ
Common questions about streaming, how the test works and how to read the results.
What is HTTPS streaming (SSE)?
A normal request waits until the server has prepared everything and returns it in one go. Streaming lets the server send while it generates, and the browser displays as it receives. SSE (Server-Sent Events) is the most common way to do this, pushing data chunk by chunk in text/event-stream format.
How does streaming relate to AI services?
AI services like ChatGPT and Claude generate answers token by token and send them as they go. If your network can't sustain tens of seconds of continuous output, answers freeze halfway or get cut off entirely.
What time to first data is normal?
This measures the network overhead between you and this site's server and doesn't include AI model thinking time. A few hundred milliseconds or less is normal; much higher usually means a long or slow proxy route.
Why do stalls happen?
Common causes are packet loss or congestion on the proxy node, route switching, middleboxes buffering responses, and an unstable device network. An occasional short stall doesn't matter much; only frequent or long stalls need fixing.
Why does it say "Disrupted"?
The page went to the background during the test. Browsers lower the priority of background pages, so the recorded arrival times aren't accurate and this result isn't used as a conclusion. Keep the page in the foreground and test again.
How much data does the test use?
The Quick tier sends 160 chunks and the Deep tier 600, each only one or two hundred bytes — a little over 100 KB at most. Results are computed only in your browser; this site doesn't store them.
Test endpoint: this site's server · /api/v1/streamStall threshold: 1.5 sec (6× the send interval)